LangChain: Managed Deep Agents — hosted runtime in LangSmith with durable execution and memory layer
Managed Deep Agents is a new LangChain hosted agent runtime published on May 13, 2026 in private beta within the LangSmith platform. The service provides durable execution, persistent memory, integrated tooling and comprehensive observability — all the infrastructure components needed for production deep agents. The agent definition stays in the repository through standard AGENTS.md and tools.json files.
This article was generated using artificial intelligence from primary sources.
LangChain introduced Managed Deep Agents on May 13, 2026 — a hosted runtime service that addresses the infrastructure complexity of production agent deployment. The product is in private beta within the LangSmith platform and was released as part of the massive coordinated release LangChain had been preparing for months.
What infrastructure problem does the service solve?
Production deep agents require four infrastructure layers that were previously independent engineering work: durable execution (survives server restarts, recovery from crashes), persistent memory (long-horizon context across days), integrated tooling (registration, versioning, sandbox for tools) and comprehensive observability (tracing, metrics, alerting). Managed Deep Agents comes with all four layers ready, eliminating weeks of setup per team.
How does the agent definition stay in the repository?
Although the runtime lives in the LangSmith cloud, the agent definition stays in the user’s repository through standard files. AGENTS.md describes the agent prompt, tool list and configuration, while tools.json contains formal tool definitions. The approach enables versioning, code review and CI/CD pipeline integration without vendor lock-in — the definition is portable and can later be migrated to a self-hosted runtime.
What does LangChain cover with this announcement?
Managed Deep Agents builds on the Deep Agents v0.6 framework (also published May 13) — an open-source agent runtime that teams can self-host. The Managed version is the “hosted compute as a service” variant for teams that want to forget about infrastructure. LangChain’s goal is to support both models: open-source for maximum control and Managed for maximum velocity.
Position in the LangSmith mass release
Its place in the broader stack: LangSmith Engine (auto debugging), LangSmith Sandboxes (test environments), LangSmith Context Hub (memory between runs), LangSmith LLM Gateway (runtime governance), SmithDB (observability data layer). Managed Deep Agents closes the loop as the deploy target — an agent created in Studio goes directly into the Managed runtime, is debugged through Engine, and evaluated through Sandbox.
Private beta access requires a LangSmith business account and contact with the LangChain sales team.
Frequently Asked Questions
- What are deep agents in the LangChain context?
- Deep agents are a LangChain class of agents designed for long-horizon tasks that require persistent state, memory hierarchy, file contexts and complex tool orchestration; the v0.6 framework version was also published on May 13 alongside the Managed runtime.
- How do Managed Deep Agents solve the infrastructure problem?
- The hosted runtime in LangSmith eliminates the need to manually set up a durable execution layer, persistent memory store, tool registry and observability stack — all come built-in, while the developer keeps the agent definition in the repository through AGENTS.md and tools.json.
Related news
arXiv:2605.22502: Compiling agentic workflows into LLM weights achieves near-frontier quality at 100× lower cost
arXiv:2605.22794: MOSS shows agents that self-improve by rewriting their own source code
arXiv:2605.22535: TerminalWorld benchmark measures LLM agents on real Linux terminal tasks without simulation